当年初DeepSeek一夜爆红，打破原有大模型市场的格局，这一年就注定不平凡。2025年的中国大模型市场经历了 一场深刻的"价值回归"，技术突破的边际效应减弱，一场围绕真实需求、可持续商业模式与产业深度的"生存进 化"全面展开。"2025年是全球化AI应用的创业之年。"顺福资本创始人、行行AI董事长李明顺总结道。 在此背景下，国内"AI六小虎"加剧赛道分化，零一万物和百川智能放弃超大模型训练，在更加务实的商业化应用 赛道越走越远，阶跃星辰将智能终端Agent作为⼤模型技术落地的关键发⼒点，在终端Agent领域取得突破，月之 暗面开始重视商业化，任命曾经的投资人为总裁，智谱和MiniMax则作为商业化的佼佼者率先成功闯关二级市 场。 DeepSeek的"起伏" 2025年初，一场由东方掀起的AI浪潮席卷全球应用市场。1月27日，来自中国的人工智能公司DeepSeek一举登顶 美国苹果商店免费应用下载榜首，将长期盘踞头部的ChatGPT暂时拉下王座，之后又迅速演变为一场全球性的现 象级传播——DeepSeek的名字随之刷屏各国社交网络，成为开年最受瞩目的科技焦点。 热度并未止步于年初的榜单登顶。整个上半年，Dee ...

文/石飞月 当年初DeepSeek一夜爆红，打破原有大模型市场的格局，这一年就注定不平凡。2025年的中国大模型 市场经历了一场深刻的"价值回归"，技术突破的边际效应减弱，一场围绕真实需求、可持续商业模式与 产业深度的"生存进化"全面展开。"2025年是全球化AI应用的创业之年。"顺福资本创始人、行行AI董事 长李明顺总结道。 在此背景下，国内"AI六小虎"加剧赛道分化，零一万物和百川智能放弃超大模型训练，在更加务实的商 业化应用赛道越走越远，阶跃星辰将智能终端Agent作为⼤模型技术落地的关键发⼒点，在终端Agent领 域取得突破，月之暗面开始重视商业化，任命曾经的投资人为总裁，智谱和MiniMax则作为商业化的佼 佼者率先成功闯关二级市场。 DeepSeek的"起伏" 2025年初，一场由东方掀起的AI浪潮席卷全球应用市场。1月27日，来自中国的人工智能公司DeepSeek 一举登顶美国苹果商店免费应用下载榜首，将长期盘踞头部的ChatGPT暂时拉下王座，之后又迅速演变 为一场全球性的现象级传播——DeepSeek的名字随之刷屏各国社交网络，成为开年最受瞩目的科技焦 点。 热度并未止步于年初的榜单登顶。整 ...

Core Insights - DeepSeek has released two new models, DeepSeek-V3.2 and DeepSeek-V3.2-Speciale, which have generated significant interest and discussion in the AI community [2][5][11] - The evolution from DeepSeek V3 to V3.2 includes various architectural improvements and the introduction of new mechanisms aimed at enhancing performance and efficiency [10][131] Release Timeline - The initial release of DeepSeek V3 in December 2024 did not create immediate buzz, but the subsequent release of the DeepSeek R1 model changed the landscape, making DeepSeek a popular alternative to proprietary models from companies like OpenAI and Google [11][14] - The release of DeepSeek V3.2-Exp in September 2025 was seen as a preparatory step for the V3.2 model, focusing on establishing the necessary infrastructure for deployment [17][49] Model Types - DeepSeek V3 was initially launched as a base model, while DeepSeek R1 was developed as a specialized reasoning model through additional training [19][20] - The trend in the industry has seen a shift from hybrid reasoning models to specialized models, with DeepSeek seemingly reversing this trend by moving from specialized (R1) to hybrid models (V3.1 and V3.2) [25] Evolution from V3 to V3.1 - DeepSeek V3 utilized a mixed expert model and multi-head latent attention (MLA) to optimize memory usage during inference [29][30] - DeepSeek R1 focused on Reinforcement Learning with Verifiable Rewards (RLVR) to enhance reasoning capabilities, particularly in tasks requiring symbolic verification [37][38] Sparse Attention Mechanism - DeepSeek V3.2-Exp introduced a non-standard sparse attention mechanism, which significantly improved efficiency in training and inference, especially in long-context scenarios [49][68] - The DeepSeek Sparse Attention (DSA) mechanism allows the model to selectively focus on relevant past tokens, reducing computational complexity from quadratic to linear [68] Self-Verification and Self-Correction - DeepSeekMath V2, released shortly before V3.2, introduced self-verification and self-correction techniques to improve the accuracy of mathematical reasoning tasks [71][72] - The self-verification process involves a verifier model that assesses the quality of generated proofs, while self-correction allows the model to iteratively improve its outputs based on feedback [78][92] DeepSeek V3.2 Architecture - DeepSeek V3.2 maintains the architecture of its predecessor, V3.2-Exp, while incorporating improvements aimed at enhancing overall model performance across various tasks, including mathematics and coding [107][110] - The model's training process has been refined to include updates to the RLVR framework, integrating new reward mechanisms for different task types [115][116] Performance Benchmarks - DeepSeek V3.2 has shown competitive performance in various benchmarks, achieving notable results in mathematical tasks and outperforming several proprietary models [127]

Core Insights - The launch of "ModelHub XC" by Paradigm Intelligence has achieved over 1,000 model adaptations within two months, four months ahead of schedule, marking significant progress in the domestic AI ecosystem [1][11] - The platform supports a diverse range of models, including general large language models, specialized vertical models, and cutting-edge innovative models, providing a solid foundation for the coordinated development of domestic AI software and hardware [1][12] Development Timeline - **Launch Date**: The platform was officially launched on September 22, 2025, addressing the compatibility issues between deployed models and underlying chip architectures, which has been a barrier to the large-scale implementation of AI [2][12] - **Vertical Model Adaptation**: On October 17, 2025, the platform completed the adaptation and deep optimization of the wind tunnel calculation model on the domestic chip Xiwang S2, achieving commercial-grade performance [4] - **Frontier Model Adaptation**: On November 1, 2025, the innovative model DeepSeek-OCR was successfully adapted for testing on various domestic computing cards, showcasing significant technical innovation [6] - **Agent Model Deployment**: On November 17, 2025, the efficient agent model MiniMax-M2 was adapted for deployment on the Ascend 910B4 chip, demonstrating global leadership in model capabilities [7] - **Batch Adaptation Achievement**: On November 25, 2025, the platform achieved large-scale adaptation of 108 models on the Moer Thread chip, highlighting its strong ecological expansion capabilities [9] Platform Capabilities - The platform is driven by the EngineX engine, enabling "plug-and-play" deployment of models on domestic chips, significantly shortening deployment cycles and resolving compatibility issues [12] - The model ecosystem is rich and diverse, covering a wide range of models and supporting major domestic computing platforms [12] - The platform offers professional services for model adaptation, backed by a team of hundreds of engineers to ensure successful adaptation and stable operation in domestic environments [12] Future Outlook - The platform aims to accelerate towards a "ten thousand model" ecosystem within a year, continuing to expand model scale and chip support [14] - The company plans to maintain a rapid update pace to build a more complete and efficient domestic AI infrastructure [14]

Core Viewpoint - Kimi, an AI startup, is focusing on open-source model development, with the recent release of Kimi K2 Thinking, which has a training cost of $4.6 million, significantly lower than competitors like DeepSeek V3 and OpenAI's GPT-3 [3][4][6] Summary by Sections Model Development and Costs - Kimi has invested heavily in open-source model research and updates over the past six months, releasing Kimi K2 Thinking on November 6, with a reported training cost of $4.6 million, lower than DeepSeek V3's $5.6 million and OpenAI GPT-3's billions [3][4] - CEO Yang Zhilin clarified that the $4.6 million figure is not official, as most expenses are on research and experimentation, making it difficult to quantify training costs [4][6] Model Performance and Challenges - Users raised concerns about the reasoning length of Kimi K2 Thinking and discrepancies between leaderboard scores and actual performance. Yang stated that the model currently prioritizes absolute performance, with plans to improve token efficiency in the future [4][7] - The gap between leaderboard performance and real-world experience is expected to diminish as the model's general capabilities improve [7] Market Position and Strategy - Chinese open-source models are increasingly being utilized in the international market, with five Chinese models appearing in the top twenty of the OpenRouter model usage rankings [7] - Kimi currently can only be accessed via API due to interface issues with the OpenRouter platform [7] - Kimi plans to maintain its open-source strategy, focusing on the application and optimization of Kimi K2 Thinking while balancing text and multimodal model development, avoiding direct competition with leading firms like OpenAI [6][8]

Core Insights - Kimi, an AI startup, has released its latest open-source model, Kimi K2 Thinking, with a reported training cost of $4.6 million, significantly lower than competitors like DeepSeek V3 at $5.6 million and OpenAI's GPT-3, which costs billions to train [1][2] - The company emphasizes ongoing model updates and improvements, focusing on absolute performance while addressing user concerns regarding inference length and performance discrepancies [1] - Kimi's strategy includes maintaining an open-source approach and advancing the Kimi K2 Thinking model while avoiding direct competition with major players like OpenAI through innovative architecture and cost control [2][4] Model Performance and Market Position - In the latest OpenRouter model usage rankings, five Chinese open-source models, including Kimi's, are among the top twenty, indicating a growing presence in the international market [2] - Kimi's current model can only be accessed via API due to platform limitations, but the team is utilizing H800 GPUs with InfiniBand technology for training, despite having fewer resources compared to U.S. high-end GPUs [2] - The company plans to balance text model development with multi-modal model advancements, aiming to establish a differentiated advantage in the AI landscape [4]

Investment Rating - The report maintains an "Overweight" rating for the technology and computer sectors [7]. Core Insights - The AI industry is entering a new paradigm characterized by the Scaling Law 2.0, where synthetic data expands the training data ceiling, and the Mid Training paradigm reshapes model evolution paths [2][3]. - The commercial application of AI is transitioning into a scaling phase, with the integration of agent capabilities and transaction loops accelerating industry implementation [2][6]. Summary by Sections Models - Computing power expansion remains the core growth engine, with representative model training computing power expected to grow at an annual rate of 4-5 times from 2010 to 2024, with leading models achieving up to 9 times [3][13]. - The cost of complete training for frontier models is projected to reach the billion-dollar level by 2027 [3][13]. Training - The Mid Training paradigm expands training boundaries by integrating reinforcement learning (RL) into the middle stage, enhancing data generation and optimal allocation [4][16]. - This approach significantly increases data utilization efficiency and is expected to break traditional performance limits [4][16]. Agents - GPT-5 establishes a "unified system" direction, promoting standardization in agent architecture through adaptive collaboration between fast and deep thinking [5][19]. - The real-time router dynamically allocates computing resources based on task complexity, enhancing response efficiency and stability in complex scenarios [5][19]. Applications - The integration of agent capabilities into commercial transactions marks a new phase of AI applications, with OpenAI's Agentic Commerce Protocol enabling AI agents to execute purchases directly [6][22]. - The global AI application landscape is evolving through three stages: productization in 2023, commercialization trials in 2024, and scaling implementation in 2025 [25][26]. - Domestic AI applications are accelerating, with significant advancements in commercial capabilities following the release of models like DeepSeek-R1 [26].

Core Insights - The article discusses an AI trading competition in the cryptocurrency market, highlighting the performance of various AI models and their strategies in a volatile environment [1][5][20]. Group 1: Competition Overview - The AI trading competition, organized by Alpha Arena, runs from October 17 to November 3, featuring real-time trading of cryptocurrencies without human intervention [1][5]. - A benchmark participant employs a simple buy-and-hold strategy for Bitcoin (BTC) to compare the performance of AI models [2]. - The competition includes a betting aspect where spectators can wager on which AI will win, adding a layer of engagement [3]. Group 2: Participating AI Models - Six leading AI models are involved: GPT-5, Gemini 2.5 Pro, Grok-4, Claude Sonnet 4.5, DeepSeek V3.1, and Qwen3 Max, each starting with $10,000 in real funds [5]. - All trades are executed on the Hyperliquid platform, ensuring transparency and security [5]. Group 3: Performance Analysis - As of October 23, Chinese models Qwen3 Max and DeepSeek V3.1 lead the competition, achieving significant profits, while Western models like GPT-5 and Gemini 2.5 Pro face substantial losses [8][10]. - Qwen3 Max adopted an aggressive strategy, leveraging high positions during market surges, resulting in a 13%-47% increase in account value [10]. - DeepSeek V3.1 maintained a steady approach, achieving 8%-21% net gains by adhering to strict risk management and diversified trading [11][12]. Group 4: Challenges Faced by Western Models - GPT-5 suffered from emotional trading and poor stop-loss management, leading to losses of 30%-40% within days, and up to 65%-75% by the end of the week [14]. - Gemini 2.5 Pro's overtrading and excessive leverage resulted in a loss exceeding 55% in the first week, highlighting the risks of high-frequency trading [14]. Group 5: Insights on Trading Strategies - Grok-4 initially gained 35% but later returned to a net loss of approximately 15% due to failure to lock in profits [15]. - Claude Sonnet 4.5, while cautious and conservative, ended with a negative return of about 17%, demonstrating the trade-off between risk and reward [19]. Group 6: Broader Implications - The competition serves as a deep experiment into the capabilities of AI in real market conditions, emphasizing that intelligence in trading is not solely about algorithmic prowess but also about adaptability in unpredictable environments [20].

Core Insights - Qwen3 Max has emerged as the leading model in the AI trading competition, surpassing DeepSeek and achieving significant profitability [1][32] - The competition, Alpha Arena, showcases the capabilities of various AI models in real market conditions, emphasizing the financial market as a training ground for AI [30][32] Performance Summary - Qwen3 Max achieved a return of +44.38%, with an account value of $14,438 and total profit of $4,438 [11] - DeepSeek V3.1 follows with a return of +20.92%, account value of $12,092, and total profit of $2,092 [11] - Other models, such as Claude 4.5 Sonnet, Grok 4, Gemini 2.5 Pro, and GPT-5, reported negative returns, with GPT-5 showing the largest loss at -71.48% [10][11] Competition Dynamics - The competition began on October 18 and has seen Qwen3 Max steadily improve its position, particularly after a significant drop in all models on October 22 [22][24] - Qwen3 Max's strategy has been characterized as "quick and precise," allowing it to capitalize on market opportunities effectively [8][32] - The competition has highlighted the contrasting performance of models, with Qwen3 Max and DeepSeek being the only two models consistently performing well [22][24] Market Implications - The success of Qwen3 Max indicates the growing competitiveness of Chinese AI models in the global market, particularly in high-risk financial environments [33] - The Alpha Arena competition serves as a demonstration of how AI can adapt and thrive in real-world financial scenarios, reinforcing the notion that financial markets are ideal for AI training [30][32]

Core Insights - A new cryptocurrency trading experiment called Alpha Arena has been launched, where leading AI models are evaluated for their investing abilities, with DeepSeek currently outperforming its competitors [1][2] - The experiment involves six large language models (LLMs) investing in cryptocurrency perpetual contracts on the decentralized exchange Hyperliquid, each starting with US$10,000 [1][2] Performance Summary - As of Tuesday, DeepSeek's V3.1 has achieved a profit of 10.11%, while OpenAI's GPT-5 has recorded losses of 39.73%, making it the worst performer [2] - Other participating models include Alibaba Cloud's Qwen 3 Max, Anthropic's Claude 4.5 Sonnet, Google DeepMind's Gemini 2.5 Pro, and xAI's Grok 4, with Grok also being a top performer [2][6] Experiment Objectives and Methodology - The primary goal of Alpha Arena is to create benchmarks that reflect real-world market dynamics, which are inherently unpredictable and adversarial [3] - The models aim to maximize risk-adjusted returns, executing trades autonomously based on shared prompts and input data, with results tracked on a public leaderboard [4] Market Engagement - DeepSeek is currently leading in prediction markets, with a 41% likelihood of topping the benchmark, and betting volume has reached US$29,707 [7] - The public can monitor trades through each model's Hyperliquid wallet address, and the reasoning behind trades is also displayed, showcasing the models' decision-making processes [4]